FPDGL: Functional-Pathological Dual Graph Learning for Alzheimer's Disease Diagnosis

• Published in: 2024 2nd International Conference on Intelligent Perception and Computer Vision (CIPCV) pp. 135 - 141
• Main Authors: Xu, Jianyu, Kang, Kai, Chen, Jilin, An, Shilong, Tang, Ruichun, Liu, Peishun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.05.2024
• Subjects: alzheimer's disease; Brain modeling; classification; Data models; GCN; Imaging; multi-modal fusion; Pathology; Predictive models; Proteins; Robustness
• DOI: 10.1109/CIPCV61763.2024.00031

In the prediction of Alzheimer's disease (AD), the widespread application of deep learning methods combining brain functional and structural imaging data is notable. Graph Convolutional Networks (GCN) have shown significant advantages in network-based brain modeling, with many studies leveraging the learning of connections between different brain regions to understand the chages in the brain. However, current research still faces challenges such as insufficient utilization of multimodal data and a lack of comprehensive modeling of the highly complex brain. To adress these issues, this paper introduces a dual graph nonlinear learning architecture (FPDGL), designed to comprehensively model the subject's brain by fully leveraging the nonlinear fusion of functional and pathological features of the brain. The article models functional and pathological networks are employed to separately learn the features of the two brain network. Subsequently, the two feature networks are fused using a dual graph nonlinear fusion approch, interaction flow of information between the two graphs, facilitating the complementary nature of multimodal features and enhancing robustness. In the classification stage, a single-layer MLP is used for the final result prediction. FPDGL fully utilizes various modal data for complex brain network modeling in predicting AD, and the obtained structure representation is termed as a comprehensive brain diagnostic entity. Based on this brain diagnostic entity, the highest achieved accuracy in the CN and AD classification tasks on the ADNI dataset is 98.36 \%.